| The wetland ecosystem is an important part of the earth’s ecosystem and has important ecological functions and values.However,with the development of social economy and the advancement of industrialization and urbanization over the years,more and more human activities have produced impacts on the wetland ecosystem.A certain degree of interference causes the already fragile wetland ecosystem to become more sensitive,and even causes degradation of wetland ecological functions such as the reduction of wetland area,the destruction of the ecological environment,and the decline of ecological functions.In this case,the wetland resources are reasonable monitoring and protection are particularly important,and wetland information extraction is the most basic and most important link to achieve monitoring and protection of wetlands.Therefore,it is very necessary to use efficient and accurate methods to classify wetland and surrounding land cover types.This article introduces the development status of wetland information extraction research at home and abroad,mainly including data sources and research methods,and also related to the development of convolutional neural networks and the Mask R-CNN method that has become popular in recent years.This paper uses Zhalong Wetland as the research area,using the remote sensing image data fused with Landsat 8 OLI optical image and Sentinel-1 radar image of Zhalong Wetland,combining the advantages of the two data sources,and obtaining a data source with richer band information select the segmented multi-source image fusion data as the classification target.With the support of software platforms such as Python3.7,Arcmap10.1,ENVI 5.3,SNAP,etc.Based on the classification method of Mask R-CNN network and segment its target into the network the research on Zhalong wetland information extraction is integrated with the local fully connected network.At the same time,the same method is used to extract wetland information from a single optical image data source and a support vector machine method to the same multi-source fusion image data source.The accuracy of the extraction results is compared and analyzed,in order to obtain higher-precision classification results of Zhalong wetland landscape types,provide more accurate methods for Zhalong wetland information extraction,enrich the research methods in the field of wetland information extraction,and realize the high efficiency of Zhalong wetland.Accurate monitoring and protection of the ecological environment provide strong support,and the following conclusions are obtained.(1)This paper uses the remote sensing image data fused with Landsat 8 OLI optical image and Sentinel-1 radar image as the data source,and uses the Mask R-CNN method of local fusion fully connected network and support vector machine method to extract information from the Zhalong Wetland.According to the comparison of extraction accuracy,it can be seen that the use of the Mask R-CNN network classification method can improve the overall accuracy of Zhalong wetland classification to a certain extent,and the classification effect is better than the commonly used SVM method.(2)This paper uses the Mask R-CNN method to extract the information of the Zhalong Wetland from the remote sensing image data and the single optical image data fused with Landsat 8 OLI optical image and Sentinel-1 radar image,combined with the comparative analysis of the extraction accuracy.The use of multi-source remote sensing image fusion data has also helped improve the classification accuracy to a certain extent.(3)The combination of Mask R-CNN and fusion data in Zhalong wetland information extraction effect is significantly better than SVM method in the extraction of fusion images and the same method in the extraction of information from a single optical data source,especially good at extracting water,farmland,and residents.These three types of land cover are a high-precision classification method for remote sensing images of Zhalong Wetland. |
PDF Full Text Request